leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 25 Packages Projects Releases Wiki Activity

v4.0 update. (#2371)

Browse Source
This commit is contained in:
Junkai-Wu
2025-06-06 02:39:20 -04:00
committed by GitHub
parent 2e2af190bd
commit 8bdbfca682
254 changed files with 29751 additions and 1980 deletions
Download Patch File Download Diff File Expand all files Collapse all files
examples/65_distributed_gemm/65_distributed_gemm.cu
+12 -10
View File
@@ -133,7 +133,7 @@ using TP = _8;
static constexpr int TP_ = TP{};
#if defined(CUTLASS_ARCH_MMA_SM90A_ENABLED) && \
(__CUDACC_VER_MAJOR__ > 12 || (__CUDACC_VER_MAJOR__ == 12 && __CUDACC_VER_MINOR__ >= 4))
(__CUDACC_VER_MAJOR__ > 12 || (__CUDACC_VER_MAJOR__ == 12 && __CUDACC_VER_MINOR__ >= 6))
// Distributed GEMM tiling/sharding schedule
// Choices:
@@ -252,7 +252,8 @@ HostTensorB tensor_B_arr[TP_];
HostTensorD tensor_C_arr[TP_];
HostTensorD tensor_D_arr[TP_];
#endif // (defined(CUTLASS_ARCH_MMA_SM90A_ENABLED) && (__CUDACC_VER_MAJOR__ >= 12) && (__CUDACC_VER_MINOR__ >= 4))
#endif // (defined(CUTLASS_ARCH_MMA_SM90A_ENABLED) &&
// (__CUDACC_VER_MAJOR__ > 12 || (__CUDACC_VER_MAJOR__ == 12 && __CUDACC_VER_MINOR__ >= 6))
/////////////////////////////////////////////////////////////////////////////////////////////////
/// Testbed utility types
@@ -345,7 +346,7 @@ struct Result {
};
#if defined(CUTLASS_ARCH_MMA_SM90A_ENABLED) && \
(__CUDACC_VER_MAJOR__ > 12 || (__CUDACC_VER_MAJOR__ == 12 && __CUDACC_VER_MINOR__ >= 4))
(__CUDACC_VER_MAJOR__ > 12 || (__CUDACC_VER_MAJOR__ == 12 && __CUDACC_VER_MINOR__ >= 6))
/////////////////////////////////////////////////////////////////////////////////////////////////
/// GEMM setup and evaluation
@@ -803,17 +804,18 @@ int run(Options &options) {
return 0;
}
#endif // (defined(CUTLASS_ARCH_MMA_SM90A_ENABLED) && (__CUDACC_VER_MAJOR__ >= 12) && (__CUDACC_VER_MINOR__ >= 4))
#endif // (defined(CUTLASS_ARCH_MMA_SM90A_ENABLED) &&
// (__CUDACC_VER_MAJOR__ > 12 || (__CUDACC_VER_MAJOR__ == 12 && __CUDACC_VER_MINOR__ >= 6))
///////////////////////////////////////////////////////////////////////////////////////////////////
int main(int argc, char const **args) {
// CUTLASS must be compiled with CUDA Toolkit 12.4 or newer to run this example
// CUTLASS must be compiled with CUDA Toolkit 12.6 or newer to run this example
// and must have compute capability at least 90.
// Some necessary cuda graph APIs were only introduced in CUDA 12.4.
if (__CUDACC_VER_MAJOR__ < 12 || (__CUDACC_VER_MAJOR__ == 12 && __CUDACC_VER_MINOR__ < 4)) {
std::cerr << "This example requires CUDA 12.4 or newer." << std::endl;
// Some necessary cuda graph APIs were only introduced in CUDA 12.6.
if (__CUDACC_VER_MAJOR__ < 12 || (__CUDACC_VER_MAJOR__ == 12 && __CUDACC_VER_MINOR__ < 6)) {
std::cerr << "This example requires CUDA 12.6 or newer." << std::endl;
// Returning zero so this test passes on older Toolkits. Its actions are no-op.
return 0;
}
@@ -857,11 +859,11 @@ int main(int argc, char const **args) {
// Evaluate CUTLASS kernels
//
#if (defined(CUTLASS_ARCH_MMA_SM90A_ENABLED) && (__CUDACC_VER_MAJOR__ >= 12) && (__CUDACC_VER_MINOR__ >= 4))
#if (defined(CUTLASS_ARCH_MMA_SM90A_ENABLED) && (__CUDACC_VER_MAJOR__ > 12 || (__CUDACC_VER_MAJOR__ == 12 && __CUDACC_VER_MINOR__ >= 6)))
run(options);
#else
std::cerr
<< "This example must be compiled with `sm90a` and CUDA Toolkit 12.4 or later." << std::endl;
<< "This example must be compiled with `sm90a` and CUDA Toolkit 12.6 or later." << std::endl;
return 0;
#endif